Search

Supplementary Data from DNA Copy Numbers Profiles in Affinity-Purified Ovarian Clear Cell Carcinoma

Supplementary Figures, Tables and Methods from Analysis of DNA Copy Number Alterations in Ovarian Serous Tumors Identifies New Molecular Genetic Changes in Low-Grade and High-Grade Carcinomas

Ovarian serous carcinoma, the most common and lethal type of ovarian cancer, is thought to develop from two distinct molecular pathways. High-grade (HG) serous carcinomas contain frequent TP53 mutations, whereas low-grade (LG) carcinomas arise from serous borderline tumors (SBT) and harbor mutations in KRAS/BRAF/ERBB2 pathway. However, the molecular alterations involved in the progression from SBT to LG carcinoma remain unknown. In addition, the extent of deletion of tumor suppressors in ovarian serous carcinomas has not been well studied. To further address these two issues, we assessed DNA copy number changes among affinity-purified tumor cells from 37 ovarian serous neoplasms including SBT, LG, and HG tumors using high-density 250K single nucleotide polymorphism arrays. Chromosomal instability index as measured by changes in DNA copy number was significantly higher in HG than in LG serous carcinomas. Hemizygous ch1p36 deletion was common in LG serous carcinomas but was rarely seen in SBT. This region contains several candidate tumor suppressors including miR-34a. In contrast, in HG serous carcinomas, significant numbers of amplifications and deletions, including homozygous deletions, were identified. Among homozygous deletions, loci containing Rb1, CDKN2A/B, CSMD1, and DOCK4 were most common, being present in 10.6%, 6.4%, 6.4%, and 4.3%, respectively, in independent 47 affinity-purified HG serous carcinomas. Except for the CDKN2A/B region, these homozygous deletions were not present in either SBT or LG tumors. Our study provides a genome-wide homozygous deletion profile in HG serous carcinomas, which can serve as a molecular foundation to study tumor suppressors in ovarian cancer. [Cancer Res 2009;69(9):4036–42]

Ovarian serous carcinoma, the most common and lethal type of ovarian cancer, is thought to develop from two distinct molecular pathways. High-grade (HG) serous carcinomas contain frequent TP53 mutations, whereas low-grade (LG) carcinomas arise from serous borderline tumors (SBT) and harbor mutations in KRAS/BRAF/ERBB2 pathway. However, the molecular alterations involved in the progression from SBT to LG carcinoma remain unknown. In addition, the extent of deletion of tumor suppressors in ovarian serous carcinomas has not been well studied. To further address these two issues, we assessed DNA copy number changes among affinity-purified tumor cells from 37 ovarian serous neoplasms including SBT, LG, and HG tumors using high-density 250K single nucleotide polymorphism arrays. Chromosomal instability index as measured by changes in DNA copy number was significantly higher in HG than in LG serous carcinomas. Hemizygous ch1p36 deletion was common in LG serous carcinomas but was rarely seen in SBT. This region contains several candidate tumor suppressors including miR-34a. In contrast, in HG serous carcinomas, significant numbers of amplifications and deletions, including homozygous deletions, were identified. Among homozygous deletions, loci containing Rb1, CDKN2A/B, CSMD1, and DOCK4 were most common, being present in 10.6%, 6.4%, 6.4%, and 4.3%, respectively, in independent 47 affinity-purified HG serous carcinomas. Except for the CDKN2A/B region, these homozygous deletions were not present in either SBT or LG tumors. Our study provides a genome-wide homozygous deletion profile in HG serous carcinomas, which can serve as a molecular foundation to study tumor suppressors in ovarian cancer. [Cancer Res 2009;69(9):4036–42]

Motivation: Multilayer perceptrons (MLP) represent one of the widely used and effective machine learning methods currently applied to diagnostic classification based on high-dimensional genomic data. Since the dimensionalities of the existing genomic data often exceed the available sample sizes by orders of magnitude, the MLP performance may degrade owing to the curse of dimensionality and over-fitting, and may not provide acceptable prediction accuracy. Results: Based on Fisher linear discriminant analysis, we designed and implemented an MLP optimization scheme for a two-layer MLP that effectively optimizes the initialization of MLP parameters and MLP architecture. The optimized MLP consistently demonstrated its ability in easing the curse of dimensionality in large microarray datasets. In comparison with a conventional MLP using random initialization, we obtained significant improvements in major performance measures including Bayes classification accuracy, convergence properties and area under the receiver operating characteristic curve (Az). Supplementary information: The Supplementary information is available on Contact: yuewang@vt.edu

The CINdex Bioconductor package addresses an important area of high-throughput genomic analysis. It calculates the chromosome instability (CIN) index, a novel measurement that quantitatively characterizes genome-wide copy number alterations (CNAs) as a measure of CIN. The advantage of this package is an ability to compare CIN index values between several groups for patients (case and control groups), which is a typical use case in translational research. The differentially changed cytobands or chromosomes can then be linked to genes located in the affected genomic regions, as well as pathways. This enables in-depth systems biology–based network analysis and assessment of the impact of CNA on various biological processes or clinical outcomes. This package was successfully applied to analysis of DNA copy number data in colorectal cancer as a part of multi-omics integrative study as well as for analysis of several other cancer types. The source code, along with an end-to-end tutorial, and example data are freely available in Bioconductor at http://bioconductor.org/packages/CINdex/ .

DNA copy number change is an important form of structural variations in human genomes. Detecting copy number changes using DNA array data is a challenging task due to high density genomic loci, low signal to noise ratios, and normal tissue contamination. We propose fused margin regression (FMR) method that combines a variable fusion rule and robust epsilon-insensitive loss criterion to approximate piecewise constant segments of the underlying copy number profile. We tested FMR on both simulation and real CGH and SNP array datasets, and observed competitively improved performance as compared to several widely-adopted existing methods.

A highly resolved tree of phenotypes (TOP) derived from genomic data reveals important relationships between heterogeneous diseases at molecular level. We propose a stability analysis guided learning method that produces a reproducible yet non-binary TOP using high-dimensional finite sample size genomic data. Experimental results show the superior capability of the proposed method in learning TOP with balanced stability and descriptiveness, as compared to conventional tree learning schemes.

Eukaryotic initiation factor 2B (eIF2B)-related disorders are heritable white matter disorders with a variable clinical phenotype (including vanishing white matter disease and ovarioleukodystrophy) and an equally heterogeneous genotype. We report 9 novel mutations in the EIF2B genes in our subject population, increasing the number of known mutations to more than 120. Using homology modeling, we have analyzed the impact of novel mutations on the 5 subunits of the eIF2B protein. Although recurrent mutations have been found at CpG dinucleotides in the EIF2B genes, the high incidence of private or low frequency mutations increases the challenge of providing rapid genetic confirmation of this disorder, and limits the application of EIF2B screening in cases of undiagnosed leukodystrophy.

For the critical task of gene module discovery in genomic research, we present a model-based hierarchical data clustering and visualization algorithm, visual statistical data analyzer (VISDA), which effectively exploits human-data interaction to improve the clustering outcome. Guided by a diagnostic tree, we apply VISDA to a muscular dystrophy dataset that contains a number of different phenotypic conditions. We then superimpose existing knowledge of gene regulation and gene function (ingenuity pathway analysis) to analyze the clustering results and generate novel hypotheses for further research on muscular dystrophies.

Though supervised and unsupervised analyses of genomic data have been intensively studied in recent years, little effort has been made to discover the structural information contained in the data. In this work, we propose a stability analysis guided supervised clustering and visualization method aiming to discover the hierarchical structure in gene expression data, which we call the "tree of phenotypes". We applied the method on two multiclass gene expression microarray data sets and presented the biological plausibility of the learned trees. We also tested the multiclass classifiers built on the learned trees and demonstrated their good classification performance.

This work presents a face detection method based on kernel Fisher discriminant analysis (KFD). Kernel based methods have been extensively investigated both in theories and applications, such as SVM and kernel PCA. Using the kernel trick, linear Fisher discriminant can be extended to non-linear case. Since the distribution of face patterns is very complex and highly nonlinear, using non-linear classification tools can hopefully tackle the problem of face detection. We explore the application of KFD in the task of frontal face detection. The experimental results prove the effectiveness of KFD in the face detection problem.

DNA copy number change is an important form of structural variations in human genomes. Detecting copy number changes using DNA array data is a challenging task due to high density genomic loci, low signal to noise ratios, and normal tissue contamination. We propose Fused Margin Regression (FMR) method that combines a variable fusion rule and robust e-insensitive loss criterion to estimate piecewise constant segments of the underlying copy number profile. We tested FMR method on both simulation and real CGH and SNP array datasets, and observed competitively improved performance as compared to several widely-adopted existing methods.